Some Best Practices in Operator Learning
This work addresses hyperparameter optimization for researchers in operator learning, but it appears incremental as it focuses on best practices without introducing new methods.
The paper tackles the problem of computationally expensive hyperparameter searches in operator learning by studying general hyperparameter and training method choices for architectures like DeepONets, Fourier neural operators, and Koopman autoencoders on differential equations, but it does not report specific numerical results.
Hyperparameters searches are computationally expensive. This paper studies some general choices of hyperparameters and training methods specifically for operator learning. It considers the architectures DeepONets, Fourier neural operators and Koopman autoencoders for several differential equations to find robust trends. Some options considered are activation functions, dropout and stochastic weight averaging.